Contents lists available at ScienceDirect

Data in Brief

Data in Brief 6 (2016) 378–385

http://d2352-34(http://c

DOIn CorrE-m

journal homepage: www.elsevier.com/locate/dib

Data Article

Feedback-related potentials in a gambling taskwith randomised reward

Faisal Mushtaq a,n, Pablo Puente Guillen a,b, Richard M. Wilkie a,Mark A. Mon-Williams a, Alexandre Schaefer c

a School of Psychology, University of Leeds, Leeds, West Yorkshire, United Kingdomb School of Computing, University of Leeds, Leeds, West Yorkshire, United Kingdomc School of Business, Monash University, Sunway Campus, Selangor, Malaysia

a r t i c l e i n f o

Article history:Received 22 October 2015Received in revised form19 November 2015Accepted 22 November 2015Available online 10 December 2015

Keywords:Event-related potentialsDecision-makingFeedback-relatedNegativityReward processingReward prediction error

x.doi.org/10.1016/j.dib.2015.11.06009/& 2015 The Authors. Published by Elsereativecommons.org/licenses/by/4.0/).

of original article: http://dx.doi.org/10.1016esponding author.ail address: f.mushtaq@leeds.ac.uk (F. Mush

a b s t r a c t

Event-related potentials (ERPs) time-locked to decision outcomesare reported. Participants engaged in a gambling task (see [1] fordetails) in which they decided between a risky and a safe option(presented as different coloured shapes) on each trial (416 in total).Each decision was associated with (fully randomised) feedbackabout the reward outcome (Win/Loss) and its magnitude (varyingas a function of decision response; 5–9 points for Risky decisionsand 1–4 points for Safe decisions). Here, we show data demon-strating: (a) the influence of Win feedback in the preceding out-come (Outcomet�1) on activity related to the current outcome(Outcomet); (b) difference wave analysis for outcome expectancy-separating Expected Outcomes (consecutive Loss trials subtractedfrom consecutive reward) from Unexpected Outcomes (subtractingLosst�1Wint trials from Wint�1Losst trials); (c) difference wavesseparating Switch and Stay responses for Outcome Expectancy;(d) the effect of magnitude induced by decisions (Riskt vs. Safet) onOutcome Expectancy; and finally, (e) expectations reflected byresponse switch direction (Risk to Safe responses vs. Safe to Riskt)on the FRN at Outcomet.

& 2015 The Authors. Published by Elsevier Inc. This is an openaccess article under the CC BY license

(http://creativecommons.org/licenses/by/4.0/).

vier Inc. This is an open access article under the CC BY license

/j.neuroimage.2015.10.046

taq).

F. Mushtaq et al. / Data in Brief 6 (2016) 378–385 379

Specifications Table

M

TH

DE

E

D

Subject area

Psychology ore specific sub-ject area

Decision-making, Cognitive Neuroscience, Reward Processing

ype of data

Event-related potentials time-locked to decision outcome. ow data wasacquired

Participants completed a two-alternative forced choice decision-making taskwhilst EEG data were acquired using a 128-channel net connected to a high-input amplifier (Electrical Geodesics, Inc., Eugene, OR; see Fig. 1 for electrodemontage) at a rate of 500 Hz (0.01–200 Hz bandwidth) and animpedancer20 kΩ for frontocentral electrodes. Data were recorded using a Czreference online and digitally converted to an average mastoids reference offline.After offline filtering (0.1–30 Hz bandwidth), data were segmented into1000 ms epochs time-locked to feedback onset (200 ms baseline) and correctedfor artifacts. Values from electrodes were clustered and mean averages fromelectrodes surrounding the standard FCz location for the FRN and P3a andelectrodes surrounding Pz for the P3 were used for statistical analysis.

ata format

Analysed ERPs. xperimentalfactors

Feedback valence was fully randomised.

xperimentalfeatures

416 trials per participant (n¼27).

ata sourcelocation

School of Psychology, University of Leeds, Leeds, United Kingdom.

ata accessibility

Within this article. D

Value of the data

� Data show feedback-locked potentials on a gambling task when outcomes are fully randomised.� The FRN, P3a and feedback-related P3 differences for outcome expectancy, switch/stay responses,

outcome magnitude and switch direction are reported.� Results demonstrate that FRN activity does not conform to existing prediction error based

accounts.

1. Data, experimental design, materials and methods

EEG data were recorded with a 128-channel net connected to a high-input amplifier (ElectricalGeodesics, Inc., Eugene, OR; for electrode montage see Fig. 1) whilst participants engaged in a two-alternative forced choice gambling task (see [1,2] for further details). On each trial, participantsdecided between a risky and a safe option, presented as different coloured shapes. In this experiment,gamble outcomes were fully randomised across 416 trials. ERPs time-locked to presentation offeedback were analysed as described in [1]. Values from electrodes were clustered and mean averagesfrom electrodes surrounding the standard FCz location for the FRN (EGI electrode numbers: ‘12’, ‘5’,‘6’, ‘13’, ‘112’, ‘7’, ‘106’, ‘Cz’, ‘31’, ‘80’ and ‘55’) and P3a and electrodes surrounding Pz for the P3 (EGIelectrode numbers: ‘61’, ‘78’, ‘62’, ‘67’, ‘72’, ‘77’, ‘71’ and ‘76’) were used for statistical analysis.

2. Reward positivity difference waves

Difference wave analysis confirmed the FRN was driven by a larger positivity to Win outcomeswhen preceded by a similarly positive outcome. We found a statistically reliable difference atOutcomet (F [1, 26]¼5.81, p¼ .023, η2p¼ .19), with Wint differences producing greater negativity

A B C D

Fig. 2. Wint and Losst FRN difference. Difference waveform computed by subtracting Wint�1Wint from Losst�1Wint trials(blue) and Losst�1�Losst from Wint�1�Losst (orange) trials at FCz (A), Cz (B) and Pz (C). Error bars represent SE and abscissashows time in milliseconds. (D) Topographical maps display mean difference at each electrode site for the FRN time window forWint and Losst (�1.1 to 0.6 mV).

Fig. 1. Electrode locations. Schematic representation of the EGI 128 HydroCelTM Geodesic Sensor Net used to acquire EEG data.

F. Mushtaq et al. / Data in Brief 6 (2016) 378–385380

(M¼�1.04 mV, SE¼ .3 mV) than Losst differences (M¼� .04 mV, SE¼ .3 mV)- complementing a sig-nificant peak-to-peak amplitude difference for the same comparison (F [1, 26]¼5.23, p¼ .031,η2p¼ .167). Commensurate with the topographical features of the FRN, Fig. 2D shows a frontocentralscalp distribution of this effect. We also performed peak-to-peak analysis to confirm that theWint�1Wint trials were also reliably different from the Losst�1Wint trials (F [1, 26]¼9.03, p¼ .006,η2p¼ .258).

A B C D

Fig. 3. Expected and Unexpected Outcome Differences. (A) FCz, (B) Cz and (C) Pz. Error bars represent SE and abscissa showstime in milliseconds. (D) Scalp maps show the Expected reward difference was maximal at frontocentral in the time-windowfor the FRN but no such pattern for the difference in Unexpected Outcomes. Maxima and minima specific to each scalp map;Expected: �2.1 to 0.6 mV; Unexpected: �2.7 to 0.69 mV.

F. Mushtaq et al. / Data in Brief 6 (2016) 378–385 381

3. Difference wave analysis: outcome expectancy

The effect of Outcome Expectancy on the FRN was examined by creating two differences waves:(i) Expected Outcome difference wave calculated by subtracting Wint�1Wint from Losst�1Losst trials;and (ii) Unexpected Outcome difference wave calculated by subtracting Losst�1Wint trials fromWint�1Losst trials. Fig. 3 shows a significant difference in FRN activity for Expectancy (F [1, 26]¼8.0,p¼ .009, η2p¼ .236), with the Expected Outcome difference producing a larger negativity(M¼�1.92 mV, SE¼ .60 mV) than Unexpected Outcome difference (M¼�0.59 mV, SE¼ .36 mV).

4. Switch vs. Stay: difference wave analysis

4.1. Switch vs. Stay: FRN

The same approach described in Section 3 was adopted for Switch and Stay trials. There was amain effect of Expectancy (F [1, 26]¼25.64, po .0001, η2p¼ .497) and the effect of Strategy approached,but did not reach, significance (F [1, 26]¼3.74, p¼ .064, η2p¼ .126). These main effects were qualified bya significant Expectancy� Strategy interaction (F [1, 26]¼27.4, po .0001, η2p¼ .513). Visually, con-sistent with Holroyd and Coles [3], the Unexpected FRN difference (M¼�1.48 mV, SE¼ .50 mV) wasmore negative than Expected (M¼�0.91 mV, SE¼ .73 mV) in Stay trials and showed a frontocentralscalp distribution (see Fig. 4B), but this effect was not statistically reliable (F [1, 26]¼0.48, p¼ .495,η2p¼ .018). The interaction was driven by Switch trials (F [1, 26]¼52.55, po .0001, η2p¼ .669). Thesetrials showed a larger FRN difference for Expected Outcome differences (M¼�5.12 mV, SE¼ .66 mV)relative to Unexpected Outcome differences (M¼ .82 mV, SE¼ .51 mV).

4.2. Switch vs. Stay: P3a

For the P3a there was a significant main effect of Expectancy (F [1, 26]¼28.91, po .001, η2p¼ .526)and Strategy (F [1, 26]¼9.11, p¼ .005, η2p¼ .259) with Switch trials eliciting a greater negativity(M¼�2.61 mV, SE¼ .43 mV) than Stay (M¼�1.11 mV, SE¼ .43 mV). These effects were qualified by anExpectancy� Strategy interaction (F [1, 26]¼46.69, po .0001, η2p¼ .642). The effect of Expectancyapproached significance in Stay trials (F [1, 26]¼3.57, p¼ .07, η2p¼ .121) with Unexpected Outcomesleading to a greater negativity (M¼�1.88 mV, SE¼ .49 mV) than Expected (M¼�0.35 mV, SE¼ .73 mV),and was significant in Switch trials (F [1, 26]¼58.83, po .0001, η2p¼ .693) with Expected Outcomesleading to a greater negativity (M¼�6.76 mV, SE¼ .78 mV) than Unexpected (M¼1.54 mV, SE¼ .59 mV).

4.3. Switch vs. Stay: feedback-related P3

Consistent with P3a results, there was a significant main effect of Expectancy (F [1, 26]¼33.86,po .0001, η2p¼ .566) and Strategy (F [1, 26]¼6.87, p¼ .014, η2p¼ .209) with Switch trials eliciting a

A

B C

Fig. 4. Switch Stay ERPs. (A) ERPs from FCz, Cz and Pz (left to right). Error bars represent SE and abscissa shows time inmilliseconds; (B) for the FRN topographical maps show Unexpected Outcomes elicited a frontocentral negativity (left) in Staytrials – consistent with Holroyd and Coles (2002), but this effect was not statistically reliable. The largest difference wasobserved for Expected Outcomes in Switch trials (maxima and minima: Stay 1.5 to �2 mV; Switch 1.4 to �5.5 mV); (C) for theP3, Unexpected Outcomes for Stay responses and Expected Outcomes for Switch responses produced posteriorly distributedfeedback-related P3 differences (maxima and minima: Stay 0.8 to �3 mV; Switch 1.8 to �3.0 mV).

F. Mushtaq et al. / Data in Brief 6 (2016) 378–385382

greater negativity (M¼�2.62 mV, SE¼ .44 mV) than Stay (M¼�1.66 mV, SE¼ .35 mV). There was asignificant Expectancy� Strategy interaction (F [1, 26]¼56.11, po .0001, η2p¼ .683). The effect ofExpectancy was significant in Stay trials (F [1, 26]¼12.95, p¼ .0013, η2p¼ .333) with UnexpectedOutcomes differences leading to less positivity (M¼�2.62 mV, SE¼ .47 mV) than Expected(M¼�0.70 mV, SE¼ .42 mV). There was also a significant effect in Switch trials (F [1, 26]¼62.31,po .0001, η2p¼ .706) with Expected Outcome differences leading to less positivity (M¼�6.01 mV,SE¼ .74 mV) than Unexpected (M¼ .78 mV, SE¼ .46 mV).

5. P3a analysis

5.1. Outcomet�2�Outcomet�1�Outcomet

There was a main effect of Outcomet (F [1, 26]¼4.28, p¼ .049, η2p¼ .141), with Wint outcomes(M¼9.72 mV, SE¼1.1 mV) relative to Losst outcomes (M¼8.78 mV, SE¼1.03 mV). No other effects orinteractions approached significance (F'so1.99, p's4 .171).

F. Mushtaq et al. / Data in Brief 6 (2016) 378–385 383

5.2. Switch vs. Stay

An Outcomet�1�Outcomet ANOVA was conducted for Stay and Switch response trials. For Stayresponses, there was no statistically reliable effects or interaction (F'so2.6, p's4 .12). For Switchtrials, there was a main effects for Outcomet (F [1, 26]¼14.43, po .0001, η2p¼ .357) with more positivegoing amplitude for Wint (M¼9.01 mV, SE¼ .96 mV) relative to Losst (M¼7.46 mV, SE¼ .88 mV) and forOutcomet�1 (F [1, 26]¼25.42, po .0001, η2p¼ .494): Wint�1 trials had a greater positive deflection(M¼9.42 mV, SE¼1.1 mV) relative to Losst�1 (M¼7.05 mV, SE¼ .72 mV). There was also anOutcomet�1�Outcomet interaction (F [1, 26]¼7.58, p¼ .011, η2p¼ .23), but this effect was qualitativelydifferent to the one observed for the FRN. This interaction was driven by differences at Outcomet�1 –

there was no difference in Outcomet for Wint�1 trials (F [1, 26]¼ .49, p¼ .49, η2p¼ .02), but a substantialeffect of Outcomet when preceded by Losst�1 (F [1, 26]¼35.60, po .001, η2p¼ .578)- Win trials fol-lowing Losses (8.37 mV, SE¼ .85 mV) were larger than sequential losses (5.74 mV, SE¼ .65 mV).

6. Risk-related ERPs

The effect of magnitude on individual ERPs was analysed by conducting anOutcomet�1�Magnitude�Outcomet ANOVA for the FRN using a peak-to-peak measure. There was asignificant main effect of Magnitude (F [1, 26]¼10.19, p¼ .004, η2p¼ .282) but this factor did notinteract with Outcomes (F’so1.36, p's4 .25). Subsequently, difference wave analyses were conductedby separating the ERPs for Outcomet for the preceding decision (Risk [producing a large outcome] vs.Safe [producing a small outcome]) and valence at Outcomet�1. Mean amplitude differences were usedin a Magnitude (Risk vs. Safe)� Expectancy (Expected Outcome Difference vs. Unexpected OutcomeDifference) ANOVA for the FRN, P3a and P3b.

6.1. FRN

There was a significant main effect of Expectancy (F [1, 26]¼9.33, p¼ .005, η2p¼ .264) – withExpected Outcomes giving rise to a greater negativity (M¼�1.99, SE¼ .52 mV) than Unexpected

A

D E F

B C

G FRN

P300

Fig. 5. Expectancy and Magnitude. (A–C) ERPs for Expected and Unexpected Reward difference waveforms from FCz, Cz and Pz(left to right) for Safe choices and (D–F) for Safe responses. Error bars represent SE and abscissa shows time in milliseconds;(G) topographical maps show the effect of Magnitude (Safe-Risk) for the FRN and P3 (maxima and minima þ1.0 mV to�2.5 mV).

F. Mushtaq et al. / Data in Brief 6 (2016) 378–385384

(M¼� .59, SE¼ .379 mV). There was also a significant main effect of Magnitude (F [1, 26]¼5.42,p¼ .028, η2p¼ .172) – with Risk choices differences leading to a larger negativity (M¼�1.89 mV,SE¼ .51 mV) than Safe (M¼� .69 mV, SE¼ .43 mV). However, Magnitude did not modulate the differ-ences in Outcome Expectancy – as indicated by the Expectancy�Magnitude interaction (F [1,26]¼ .015, p¼ .903, η2p¼ .001). To illustrate, difference waveforms for Risk and Safe are plotted sepa-rately in Fig. 5.

6.2. P3a

There was a significant main effect of Magnitude (F [1, 26]¼13.58, p¼ .001, η2p¼ .343) – with Riskchoices leading to a larger negativity (M¼�2.3 mV, SE¼ .52 mV) than safe (M¼� .52 mV, SE¼ .44 mV).However, there was no effect of Expectancy (F [1, 26]¼2.78, p¼ .107, η2p¼ .097) and no Expectan-cy�Magnitude interaction (F [1, 26]¼ .08, p¼ .783, η2p¼ .003).

6.3. Feedback-related P3

There was a significant main effect of Magnitude (F [1, 26]¼11.48, p¼ .002, η2p¼ .306) – with Riskchoices leading to a larger negativity (M¼�2.42 mV, SE¼ .33 mV) than Safe (M¼�1.35 mV,SE¼ .33 mV). However, there was no effect of Expectancy (F [1, 26]¼1.50, p¼ .232, η2p¼ .054) and theExpectancy�Magnitude interaction did not reach significance (F [1, 26]¼1.38, p¼ .251, η2p¼ .05).

7. Switch direction�Outcomet analysis

An Outcomet�1�Outcomet� Switch Direction (Riskt�1Safet; Safet�1Riskt) ANOVA was not pos-sible due to a constraint with the number of trials available for this analysis (only 3 subjects had 416trials across the 8 conditions). In order to examine the impact of expectancy reflected by Switchsequence direction on Outcomet related activity a Switch Direction�Outcomet ANOVA was con-ducted. All 27 subjects had a minimum of 16 artefact-free trials available for this analysis: Riskt�1SafetLosst (trials M¼42, SD¼11); Safet�1Riskt Losst (trials M¼40, SD¼9); Riskt�1Safet Wint (trials M¼39,SD¼8); Safet�1Riskt Wint (trials M¼42, SD¼9).

There was a main effect of Outcomet (F [1, 26]¼7.05, p¼ .013, η2p¼ .213) and Switch Direction (F [1,26]¼5.38, p¼ .029, η2p¼ .171), with Riskt�1Safet (M¼�4.45 mV, SE¼ .72 mV; see Fig. 6A) more negativethan Safet�1Riskt (M¼�3.79 mV, SE¼ .70 mV). Although visually, the ERPs indicated differences(Fig. 6B), there was no statistically reliable Outcomet� Switch Direction interaction (F [1, 26]¼ .01,p¼ .919, η2po .001).

A B

Fig. 6. Switch direction and outcome related activity. (A) ERPs at FCz showed marginally significant peak-to-peak FRN effectwith greater negativity for Riskt�1Safet responses relative to Safet�1Riskt, however, there was no statistically reliable SwitchDirection�Outcome interaction. (B) Abscissa represents time in milliseconds.

F. Mushtaq et al. / Data in Brief 6 (2016) 378–385 385

Acknowledgements

This research was supported by the Biotechnology and Biological Sciences Research Council(BBSRC; Grant number BB/H001476).

References

[1] F. Mushtaq, R.W. Wilkie, M. Mon-Williams, A. Schaefer, Randomised prior feedback modulates neural signals of outcomemonitoring, NeuroImage 125 (2016) 868–879. http://dx.doi.org/10.1016/j.neuroimage.2015.10.046.

[2] F. Mushtaq, G. Stoet, A.R. Bland, A. Schaefer, Relative changes from prior reward contingencies can constrain brain corre-lates of outcome monitoring, PloS One 8 (2013) e66350. http://dx.doi.org/10.1371/journal.pone.0066350.

[3] C.B. Holroyd, M.G.H. Coles, The neural basis of human error processing: reinforcement learning, dopamine, and the error-related negativity, Psychol. Rev. 109 (4) (2002) 679–709. http://dx.doi.org/10.1037//0033-295X.109.4.679.